DocDrift

The problem: code changes every week, but READMEs and setup guides do not. Files get renamed, routes move, settings change name, and nothing checks the docs against the code. A new developer follows the README, step 3 fails, and an hour is gone. After a few of those, nobody trusts the docs. The solution: DocDrift treats every checkable sentence in the documentation as a claim and proves or disproves it against the actual code. One command in IBM Bob, /docdrift <repo>, clones the repository, checks its licence, extracts the claims (commands, file paths, settings, API routes, versions, dependencies, counts) and checks them in parallel. Every problem must cite a file and line in the repository and is double-checked before it is reported. The result is a report with a trust score and, for each finding, what the docs say next to what the code says. You then choose what to fix. The fixes are made to the documentation only, on a separate branch, and nothing is committed without your approval. DocDrift never pushes. Target users: onboarding and platform teams, open-source maintainers, and teams whose documentation must be accurate. They use it in Bob IDE with one slash command, or from a dashboard where they paste a repository link, click Audit and see the report appear, then copy a ready-made fix command back into Bob. What makes it different: it is measured and honest. On a README with 10 planted errors, from a missing file to a wrong HTTP method, DocDrift caught 10 of 10 with 0 false alarms in about 3 minutes for 1.65 Bobcoins. In my own project it found a HIGH problem (the README promised a SQLite setup but the example settings file was missing); after the fix, following the README from scratch passed 44 tests and a re-audit scored 100%. On unfamiliar MIT projects it found a real stale CI badge and reported correct READMEs as 100% without inventing problems. Every mistake it made during the hackathon became a rule: it now says UNVERIFIED instead of guessing, treats outside knowledge as advice rather than drift, searches documentation inside code before calling anything missing, and takes facts like the date from tools. One audit costs under 2 Bobcoins. Links: live dashboard with real reports at bekarys2323322332.github.io/docdrift/dashboard/ and the full build log, benchmark answer key and data sources in the GitHub repository.